Research On Civil Aviation Aircraft Trajectory Prediction Based On ADS-B Data

With the rapid development of my country’s air transport industry,the number of flights is increasing,which poses a huge challenge to Air Traffic Management(ATM).In this context,the International Civil Aviation Organization(ICAO)has adopted Trajectory Based Operation(TBO)as the core operating concept of the next-generation air navigation system.TBO takes the flight trajectory of the aircraft as the only reference,and realizes the sharing of flight trajectories in the ATM system.All parties make collaborative decisions and accurately control the operation of the aircraft.As the main means of monitoring part of the traditional air traffic control system,radar is mainly composed of primary radar and secondary radar system.The Automatic Dependent Surveillance-Broadcast(ADS-B)system is an auxiliary surveillance method that relies on the satellite-based global positioning system and the transmission of the ground-air data link.It has high positioning accuracy and low maintenance cost,and is considered to be a promising technology in the new generation of ATM systems.In recent years,with the continuous development of artificial intelligence,the accuracy of trajectory prediction,real-time performance,robustness of neural network models and lightweight requirements have been continuously improved.Given that ADS-B data is a multidimensional time series with rich spatiotemporal features,it has strong complexity and uncertainty.In this paper,trajectory prediction methods based on deep learning are studied in depth.The main research results are as follows:(1)Research based on Long Short-Term Memory(LSTM),Bi-directional Long Short-Term Memory(Bi-directional LSTM),Stacked Long Short-Term Memory(Stacked LSTM)trajectory prediction method.Considering that the ADS-B trajectory data is a multivariate time series,the Recurrent Neural Network(RNN)can well capture the characteristics of the trajectory in the time dimension.This paper firstly designs the LSTM prediction model and the Bi-directional LSTM prediction model,and conducts experiments and performance analysis in the case of univariate,single step and short time.It is found that the overall performance of the LSTM prediction model is better than the Bi-directional prediction model and the prediction error of the LSTM model is reduced on average.2.94%.Then,by building a Stacked LSTM prediction model with multiple hidden LSTM layers,and comparing the performance with the LSTM prediction model,it is found that the performance of the Stacked LSTM prediction model has not improved,but has declined,and the prediction error has increased by 37.82% on average.(2)The trajectory prediction method based on CNN-LSTM is studied.Considering the spatial features of ADS-B trajectory data,Convolutional Neural Networks(CNN)are more suitable for extracting spatial features.In this paper,a CNN-LSTM composite model is designed,and experiments are also carried out in the case of univariate single-step short time.It is found that the overall prediction performance of the CNN-LSTM composite model is better than that of the LSTM model,and the average prediction error of the CNN-LSTM model is reduced by 52.68%.(3)The trajectory prediction method based on Mobile Net-LSTM is studied.Considering that CNN is mainly used in the field of computer vision,the model is complex and the amount of calculation is large,it is not applicable in some real scenes that require low latency and fast response speed.In this paper,a lightweight Mobile Net-LSTM composite model is designed,and experiments are also carried out in the case of univariate single step and short time.It is found that the performance of Mobile Net-LSTM is lower than that of CNN-LSTM,but it is better than the LSTM model.Compared with the LSTM model,the Mobile Net-LSTM prediction error is reduced by an average of 35.77%,and compared with the CNN-LSTM model,the Mobile Net-LSTM prediction error is increased by an average of 35.73%,but it provides options for trajectory prediction in more real-world scenarios,which has a profound impact on intelligent air traffic control.